Two-Domain DNA Strand Displacement

Two-Domain DNA Strand Displacement Luca Cardelli Microsoft Research Cambridge, UK [email protected] We investigate the computing power of a restricted class of DNA strand displacement structures: those that are made of double strands with nicks (interruptions) in the top strand. To preserve this structural invariant, we impose restrictions on the single strands they interact with: we consider only two-domain single strands consisting of one toehold domain and one recognition domain. We study fork and join signal-processing gates based on these structures, and we show that these systems are amenable to formalization and to mechanical verification. 1 Introduction Among the many techniques being developed for molecular computing [5], DNA strand displacement has been proposed as mechanism for performing computation with DNA strands [8, 3]. In most schemes, single-stranded DNA acts as signals and double-stranded (or more complex) DNA structures act as gates. Various circuits have been demonstrated experimentally [10]. The strand displacement mechanism is appealing because it is autonomous [4]: once signals and gates are mixed together, computation proceeds on its own without further intervention until the gates or signals are depleted (output is often read by fluorescence). The energy for computation is provided by the gate structures themselves, which are turned into inactive waste in the process. Moreover, the mechanism requires only DNA molecules: no organic sources, enzymes, or transcription/translation ingredients are required, and the whole apparatus can be chemically synthesized and run in basic wet labs. The main aims of this approach are to harness computational mechanisms that can operate at the molecular level and produce nano-scale structures under program control, and somewhat separately that can intrinsically interface to biological entities [2]. The computational structures that one may easily implement this way (without some form of unbounded storage) vary from Boolean networks, to state machines, to Petri nets. The last two are particularly interesting because they take advantage of DNA’s ability to encode symbolic information: they operate on DNA strands that represent abstract signals. The fundamental mechanism in many of these schemes is toehold mediated branch migration and strand displacement [10], which implements a basic step of computation. It operates as shown in Figure 1, where each letter and corresponding segment represents a DNA domain (a sequence of nucleotides, C,G,T,A) and each DNA strand is seen as the concatenation of multiple domains. Single strands have an orientation; double strands are composed of two single strands with opposite orientation, where the bottom strand is the Watson-Crick, C − G, T − A, complement of the top strand. The ‘short’ domains hybridize (bind) reversibly to their complements, while the ‘long’ domains hybridize irreversibly; the exact critical length depends on physical condition. Distinct letters indicate domains that do not hybridize with each other. In the first reaction of Figure 1, a short toehold domain t initiates binding between a double strand and a single strand. After the (reversible) binding of the toehold, the x domain of the single strand gradually replaces the top x strand of the double strand by branch migration. The branching point between the S. B. Cooper, E. Kashefi, P. Panangaden (Eds.): Developments in Computational Models (DCM 2010) EPTCS 26, 2010, pp. 47–61, doi:10.4204/EPTCS.26.5 48 Two-Domain DNA Strand Displacement Figure 1: Toehold-mediated DNA branch migration and strand displacement = Figure 2: Examples of allowable single and double strands: tyxytyxyt;tx;xt;x two top x domains performs a random walk that eventually leads to displacing the x strand. The final detachment of the top x strand makes the whole process essentially irreversible, because there is no toehold for the reverse reaction. The second reaction illustrates the case where the top domains do not match: then the toehold binds reversibly and no displacement occurs. The third reaction illustrates the more detailed situation where the top domains matches only initially: the branch migration can proceed only up to a certain point and then must revert back to the toehold: hence no displacement occurs and the whole reaction reverts. The fourth reaction illustrates a toehold exchange, where a branch migration (of strand tx) leads to a displacement (of strand xt), but where the whole process is reversible via a reverse toehold binding and branch migration. The first (irreversible) and fourth (reversible) reactions are the fundamental steps that can be composed to achieve computation by strand displacement. 2 Two-domain Signals and Gates We now describe some DNA strand displacement structures that emulate, depending on the point of view, either chemical reactions or Petri net transitions. Their function is to join input signals and fork output signals. To achieve compositionality, so that gates can be composed arbitrarily into larger circuits, it is necessary to first fix the structure of the signals. Any given choice of signal structure requires a different gate architecture, for example for 4-domain signals [9] (signals composed of 4 segments of different function), and 3-domain signals [1]. Here we present a new, streamlined, architecture based on 2-domain signals, where the gates can be combined into arbitrary circuits (including loops), and where the waste products do not interfere with the active gates. Luca Cardelli 49 Figure 3: Transducer Txy j tx ! ty: initial state plus input tx. Top-nicked double strands. Double-stranded DNA (dsDNA) can have interruptions (nicks) on one strand while remaining connected if the opposite strand has enough hold on the area around the nick. We called such structures nicked double-stranded DNA (ndsDNA). This excludes any long overhangs or any protrusions from the double- strand. In particular, we work with top-nicked double-strands, where all the nicks are on one strand (the top one by convention). A deviation from this simple structure happens fleetingly during branch migration, but all the initial and final species we use are ndsDNA. We use t for short domains, x,y,z for long domains, and a,b,c for long domains that are meant to be privately used by some construction. We write, e.g., tx for a single-stranded DNA (ssDNA) strand consisting of a toehold t followed by a domain x, and similarly for xt. We write, e.g., txy for a fully complemented double strand consisting of a continuous strand txy at the top and its Watson-Crick complement at the bottom. Finally, we write txyy to indicate the same double strand but with a nick at the top between x and y. In the figures, a nick is indicated by an arrowhead and a discontinuity. Examples of allowable single and double strands are shown in Figure 2. We assume that domains indicated by different letters are distinct, so that, e.g., x does not hybridize with y, zy, yz, ty, or yt. To simplify our notation, we use an implicit equivalence illustrated in the bottom part of the figure. Suppose we start with a regular double strand, and we nick it at the top (bottom left). Long segments between nicks remain attached to the bottom strand, while short toehold segments can detach and reattach (bottom right). We regard these reversible states as equivalent; the notation xytyy then indicates two equivalent situations, where the top t is either present or absent, and where t is implicitly exchanged with the environment. Hence, we can use xytyy to indicate an open toehold between x and y, because the toehold is available (sometime). This way, we do not need to use separate notations for temporarily occluded and temporarily open toeholds, which we would have to regard as equivalent anyway (up to some kinetic occlusion effects). Two-domain strand displacement gates. All our gates are top-nicked dsDNA and our signals are two-domain ssDNA. This simple setup is more expressive than it might appear at first. For example (Figure 3), let us consider a single strands tx as encoding a signal, with the strand xt as its cosignal, and consider the problem of constructing a signal transducer Txy from a signal tx to a signal ty, with the reduction Txy j tx ! ty, where j is parallel composition of components, and final waste is discarded. All signals share the same toehold t, and are distinguished by the long domains x,y,z, etc. As shown in Figure 4, the input tx can initiate a signal/cosignal cascade of strand displacements in the left double-strand that after two toehold exchanges releases a private cosignal at (the segment a is privately used by the Txy transducer, with a distinct a for each xy pair). The at cosignal then initiates a backward cascade in the right double strand that releases the desired output signal ty at the fourth reaction. The release of ty is reversible, but the gate is then 50 Two-Domain DNA Strand Displacement Figure 4: Transducer Txy j tx ! ty reactions. Figure 5: Fork Fxyz j tx ! ty j tz: initial state plus input tx. locked down by the last two reactions. The locking down of the gate is also used to reabsorb the xt and ta strands, by exploiting the x end of the right structure and the a end of the left structure. In the end, only unreactive (no exposed toeholds) dsDNA and ssDNA is left. In Figure 4, the initial structures from Figure 3 are shown inside rounded rectangles, and the final structures inside squared rectangles. The reaction rules are described abstractly in Figure 10. y y y The structures in Figure 3 can be written in the notation described above as Txy = t xt at a j ta j xytyytayt j yt.

Load more